| v6.8.0 | ### Added - **`contribai prs` command** โ list submitted PRs from local memory with status filter: - `--status open|merged|closed|failed|all` (default `all`) - `--limit N` (default 20) - `--json` โ emit a JSON array (PR number parsed as int) for piping to `jq`/scripts - Pretty mode color-codes status (green=merged, cyan=open, red=closed, yellow=failed) and renders date / PR# / repo / title / URL. - Complements `stats` (which shows only the last 5) by giving a full filtered view. - 4 u | High | 4/27/2026 |
| v6.4.1 | See [CHANGELOG.md](https://github.com/tang-vu/ContribAI/blob/main/CHANGELOG.md) for details. | High | 4/14/2026 |
| v6.4.0 | See [CHANGELOG.md](https://github.com/tang-vu/ContribAI/blob/main/CHANGELOG.md) for details. | High | 4/14/2026 |
| v6.3.0 | See [CHANGELOG.md](https://github.com/tang-vu/ContribAI/blob/main/CHANGELOG.md) for details. | Medium | 4/14/2026 |
| v6.2.1 | See [CHANGELOG.md](https://github.com/tang-vu/ContribAI/blob/main/CHANGELOG.md) for details. | High | 4/14/2026 |
| v6.2.0 | ### Added - **Sprint 18 complete** โ Dependencies, benchmarks, binary optimization: - Criterion benchmark suite (5 benchmarks: AST extraction, framework detection, risk classification) - Rust dependabot (weekly automated dependency updates) - Test fixtures for benchmarking (Python/Rust/JavaScript samples) ### Changed - **Dependencies updated**: - tower 0.4 โ 0.5 (compatible with axum 0.7) - Tree-sitter grammar audit (documented 0.23/0.24/0.25 compatibility) - **Binary size optimizati | High | 4/10/2026 |
| v6.1.0 | ### Added - **Sprint 17 complete** โ Code quality & dead code removal: - Framework detection from imports (20+ frameworks: Django, React, Rails, etc.) - Copilot provider fully wired in all factory functions - Session dead code removed (commented out for future feature) - 5 new framework detection tests ### Fixed - **6 clippy warnings eliminated** โ Zero-warning strict lint: - Removed unused imports (`OpenOptions`, `std::io::Write`, `warn`, `CopilotProvider`) - Fixed `unwrap()` afte | Medium | 4/10/2026 |
| v6.0.0 | ## Plugin System + Enterprise Mode + i18n **BREAKING CHANGE**: Major version bump for plugin system architecture and enterprise features. ### Plugin System Extensible lifecycle hooks for pipeline events: ```yaml plugins: - name: "slack-notifier" hooks: [on_pr_created, on_error] config: webhook_url: "https://hooks.slack.com/test" channel: "#devops" - name: "custom-audit" hooks: [on_analysis_complete, on_pr_merged] ``` **Available hooks:** - `on_analysis_complete` โ | High | 4/7/2026 |
| v5.20.0 | ## Client/Server Architecture + TUI Polish + Observability Complete infrastructure upgrade: remote API access, session management, and structured logging. ### Pipeline Server Mode - **`contribai serve`** โ starts pipeline server with REST API (default port: 9876) - Bind to custom host/port: `contribai serve --host 0.0.0.0 --port 9876` ### REST API Endpoints | Endpoint | Method | Auth | Description | |----------|--------|------|-------------| | `/api/health` | GET | Optional | Server health ch | Medium | 4/7/2026 |
| v5.19.0 | ## Agent Modes + Permissions + Small Model + Sessions Complete feature pack: read-only analysis, rule-based access control, cost optimization, and undo support. ### Agent Modes - **`plan` mode**: read-only analysis, no code generation or PRs - **`build` mode**: full PR flow (default) - **CLI**: `contribai run --mode plan` ### Rule-Based Permission System - Granular file/shell access control with glob patterns (`*`, `**`, `?`) - Actions: `allow` | `ask` | `deny` - 6 permission types: `file_rea | Medium | 4/7/2026 |
| v5.18.0 | ## Sprint 11: Auth Ecosystem GitHub Copilot, Auto-Detection, Token Refresh, Fallback Chain. ### GitHub Copilot Provider - Exchange `gh auth` token for Copilot API access - Access **gpt-4o**, **claude-sonnet-4**, **gemini-2.5-pro** via Copilot subscription - No separate API key needed โ just `gh auth login` - Auto-refresh token before expiry (5-min TTL, 30s safety buffer) ### Enhanced Login Flow - `contribai login` now detects **ALL** available auth sources: - **Copilot** (via gh CLI) - ** | Medium | 4/7/2026 |
| v5.17.1 | ## Fix: Config Loading Bug ### Root Cause - `load_config(None)` was returning `ContribAIConfig::default()` instead of calling `ContribAIConfig::load()` - This caused **ALL** commands (`run`, `login`, etc.) to ignore user's saved `config.yaml` - Result: always falling back to `gemini` defaults, even when user configured `vertex` or `copilot` ### Fix - `load_config(None)` now searches default locations in order: 1. `./config.yaml` 2. `./config.yml` 3. `~/.contribai/config.yaml` 4. Defaul | Medium | 4/7/2026 |
| v5.17.0 | ## Sprint 1-7 Complete โ First Pre-Built Release All 7 sprints combined into one major release with pre-built binaries for 4 platforms. ### Sprint 1: Critical Reliability (v5.9.0) - Circuit breaker for LLM failures (Closed โ Open โ HalfOpen) - E2E LLM parser tests against 18 real response shapes - Analyzer retry with exponential backoff (2s/4s/8s) - **Tests**: +51 ### Sprint 2: Security Hardening (v5.10.0) - Sandbox enabled by default (`sandbox.enabled: true`) - Prompt injection protection (1 | Medium | 4/7/2026 |
| v5.16.0 | ## Documentation + Observability ### Documentation - **`ARCHITECTURE.md`**: Comprehensive system documentation - System overview, pipeline flow, component map - Data model, config reference, how-to guides - Adding CLI commands, LLM providers, analysis skills - **`RUNBOOK.md`**: Troubleshooting guide - 12 common issues with step-by-step resolutions - Debug mode, log locations, maintenance tasks - Emergency procedures (stop all runs, reset circuit breaker, nuclear reset) ### Observab | Medium | 4/7/2026 |
| v5.15.0 | ## Web Dashboard TLS Support ### HTTPS Dashboard Enable TLS via config: ```yaml web: tls_enabled: true tls_cert_path: /path/to/cert.pem tls_key_path: /path/to/key.pem ``` - Uses `axum-server 0.7` + `rustls` (no OpenSSL dependency) - Dashboard URL automatically switches from `http://` to `https://` - Config validation: rejects missing cert/key paths - Config validation: rejects `0.0.0.0` binding without TLS ### Dependencies Added - `axum-server 0.7` with `tls-rustls` feature ### Stats - | Medium | 4/7/2026 |
| v5.14.0 | ## Dream Race Condition Fix ### TOCTOU Race Fix - Replaced string-based DB lock with `DreamLock` using `Mutex<()>` - Eliminates race condition where two concurrent `maybe_dream()` calls could both pass gate checks and run consolidation simultaneously - **Root cause**: Old lock used a DB field (`dream_lock = "1"`) with check-then-set pattern โ two instances could both see `0`, both set to `1`, both run ### DreamLock Implementation - Mutex-based advisory lock prevents concurrent dream consolidat | Medium | 4/7/2026 |
| v5.13.0 | ## Test Coverage Sprint ### 67 New Tests Across 5 Test Suites **`ast_all_languages.rs`** (22 tests) - Symbol extraction for all 13 languages: Python, JS, TS, Go, Rust, Java, C, C++, Ruby, PHP, C#, HTML, CSS - Edge cases: empty files, syntax errors, unicode content, deeply nested structures (50 levels), mixed file extensions **`middleware_chain.rs`** (11 tests) - Rate limit enforcement (allows under limit, blocks at limit, allows in dry-run) - Validation (passes valid repo, blocks empty repo n | Medium | 4/7/2026 |
| v5.12.0 | ### Added - **LLM response cache**: Content-addressable SHA-256 cache for `complete()` calls. Identical prompts return cached responses, skipping the API. Configurable TTL (default: 7 days) via `llm.cache_enabled` and `llm.cache_ttl_days`. - **Parallel file fetching**: Analyzer now fetches file contents concurrently (10 concurrent requests via semaphore), reducing analysis time by ~60% for large repos. - **`contribai cache-stats`**: Shows cache size, valid/expired entry counts, hit rate. - **`c | Medium | 4/6/2026 |
| v5.11.0 | ### Added - **CLI refactored**: `cli/mod.rs` reduced from 3,396 lines โ **523 lines** (85% reduction). All 26 command handlers extracted into `cli/commands/` directory, shared utilities in `cli/common.rs`. - **Magic numbers extracted** to config fields (`SandboxConfig.mode`, `require_validation`). - **Expanded local validators**: JavaScript/TypeScript, Go, Java syntax checks added. ### Refactored - `cli/mod.rs`: 3,396 โ 523 lines - Each command handler in its own file under `cli/commands/` - S | Medium | 4/6/2026 |
| v5.10.0 | ### Added - **Sandbox enabled by default** (`sandbox.enabled = true`) with 3 modes: `"docker"` (full isolation), `"local"` (syntax check, default), `"ast"` (tree-sitter parse), `"off"` (no validation). `sandbox.require_validation` blocks PR submission on validation failure. - **Prompt injection protection** โ repository content is sanitized before LLM calls: control characters stripped, XML-wrapped in `<repository-content>` tags, 10 known injection patterns detected and logged. System prompts h | Medium | 4/6/2026 |
| v5.9.0 | ### Added - **Circuit Breaker for LLM failures**: Full circuit breaker pattern (Closed โ Open โ HalfOpen) stops pipeline after consecutive LLM failures to save API quota. Configurable thresholds via `pipeline.circuit_breaker_failure_threshold` (default: 5), `success_threshold` (default: 2), `cooldown_secs` (default: 300). - **`contribai circuit-breaker` CLI command**: Shows current circuit state, failure count, cooldown remaining, and recovery status. - **Analyzer retry with exponential backoff | Medium | 4/6/2026 |
| v5.8.1 | ## What's New ### ๐ Resolved Imports Separation (`core/models.rs`, `orchestrator/pipeline.rs`) - New `resolved_imports` field cleanly separates cross-file import data from `symbol_map` - Updated 6 construction sites across pipeline, engine, and models ### ๐ก๏ธ AST Depth Guard (`analysis/ast_intel.rs`) - Max depth 8 on `walk_import_nodes` to prevent infinite recursion in circular imports ### ๐ Closed-PR Failure Analysis (`pr/patrol.rs`) - `analyze_closed_pr()` fetches reviewer feedback from r | Medium | 4/5/2026 |
| v5.8.0 | ## What's New ### ๐ Cross-file Import Resolution (`analysis/ast_intel.rs`) - **5 languages**: Python, JS/TS, Go, Rust, Java import graph traversal - 1-hop resolution with 20-file cap to prevent explosion - `symbol_map` now properly wired in pipeline (was always empty before) - Enables understanding of cross-module dependencies for smarter fixes ### ๐งช Integration Test Suite - **9 new integration tests** (5 patrol + 4 hunt scenarios) - Mock GitHub infrastructure via `GitHubClient::with_base_ur | Medium | 4/5/2026 |
| v5.7.1 | ## What's New ### ๐ง Bug Fixes - **Hunt command**: CLI `hunt` now correctly calls `pipeline.hunt()` instead of `pipeline.run()`, enabling multi-round aggressive discovery mode - **Code formatting**: `cargo fmt --all` applied across entire codebase for consistent style ### ๐ Stats - All 394 tests passing, 0 failures **Full Changelog**: https://github.com/tang-vu/ContribAI/compare/v5.7.0...v5.7.1 | Medium | 4/4/2026 |
| v5.7.0 | ## What's New ### โณ LLM Retry with Exponential Backoff (`llm/retry.rs`) - **RetryingProvider** wrapper wraps all LLM providers automatically - Retries on rate limits (429), server errors (5xx), and network timeouts - Does NOT retry auth errors โ fails immediately for misconfig - 3 retries, 1s base delay, exponential growth (1โ2โ4s), capped at 30s - Jitter prevents thundering herd on shared API keys - `create_llm_provider()` now wraps every provider with retry - `create_llm_provider_raw | Medium | 4/4/2026 |
| v5.6.0 | ## What's New ### ๐ง Semantic Context Compression (`analysis/compressor.rs`) - **3-tier compression**: Full source โ Signature extraction โ Truncated middle - **Semantic chunking**: Splits code at function/class boundaries using AST symbols - **Signature extraction**: Strips implementation bodies for Python, JS/TS, Rust, Go, Java - Default token budget: 30k tokens ### ๐ฌ PR Conversation Memory (`orchestrator/memory.rs`) - New `pr_conversations` SQLite table stores full PR interaction | Medium | 4/4/2026 |
| v5.5.0 | ## What's New in v5.5.0 ### Multi-file PR Batching Pipeline now merges related findings into a **single multi-file PR** instead of creating separate PRs per finding. PRs with 2-5 file changes look more human and are more likely to be merged. ### Issue Solver (End-to-End) `contribai solve <repo_url>` now actually works โ fetches solvable issues, generates code fixes via LLM, and creates PRs with `Fixes #N` auto-close linking. Previously only listed issues. ### PR Conversation Memory Patrol sto | Medium | 4/3/2026 |
| v5.4.2 | See [CHANGELOG.md](https://github.com/tang-vu/ContribAI/blob/main/CHANGELOG.md) for details. | Medium | 4/3/2026 |
| v5.4.1 | # ContribAI v5.4.1 ## ๐ง Bug Fixes ### MCP Server: Fix `invalid character '_'` initialization error The MCP JSON-RPC protocol communicates over **stdout**. Two sources were corrupting the channel: 1. **`tracing_subscriber`** defaulted to writing log output to stdout 2. **`print_banner()`** + `println!()` wrote ASCII art and status messages to stdout Clients (Claude Desktop, Antigravity) would receive `_____ ContribAI...` before the JSON response, causing `invalid character '_' looking for b | Medium | 4/2/2026 |
| v5.4.0 | # ContribAI v5.4.0 ## ๐ Dream Memory Consolidation New `dream` system that consolidates scattered memory entries into durable, well-organized repo profiles. - **3-gate trigger**: Runs automatically when 24h have passed + 5 sessions completed + no concurrent dream - **Repo profiling**: Aggregates PR outcomes (merged/closed), feedback patterns, and review times into personality profiles per repo - **Smarter targeting**: Dream profiles inform which contribution types to prioritize for each repo | Medium | 4/2/2026 |
| v5.3.0 | ๏ปฟ## What's New in v5.3.0 ### 13-Language Deep AST Analysis (tree-sitter) Expanded from 8 to **13 languages** with native tree-sitter parsing: - **New**: Ruby, PHP, C#, HTML, CSS - **Fallback**: Kotlin/Swift use Java parser, Vue/Svelte use HTML parser ### All-Language Discovery (default) - Default now searches **all 15 languages** (was: python only) - Narrow via: contribai config-set discovery.languages ### Fixed: target and analyze Commands - Now **directly target** the specified repo URL - P | Medium | 4/1/2026 |
| v5.2.1 | ๏ปฟ## What's Changed in v5.2.1 ### Critical Bug Fixes - **Vertex AI endpoint**: Fixed 404 error for global location - **target and analyze commands**: Fixed - now directly targets the specified repo - **Preview models**: v1beta1 API version for -preview models - **Tilde path expansion**: Fixed ~/.contribai/memory.db ### New Features - Gemini 3.x model catalog (3.1-pro, 3-pro, 3-flash, 3.1-flash-lite) - run_targeted() pipeline method - Better error logging with response preview - Default model: | Medium | 4/1/2026 |
| v5.2.0 | ๏ปฟ# v5.2.0 ## What's New - Interactive Login (contribai login) - full config menu with dialoguer - Switch LLM providers (Gemini, Vertex AI, OpenAI, Anthropic, Ollama) - 4-Platform Binaries: Linux x86_64, Windows x86_64, macOS Intel, macOS ARM64 - Rust-First CI: fmt + clippy -D warnings + 335 tests + audit - Python legacy tests only with [python] label ## Fixes - Resolved 24+ clippy warnings across 9 files - Replaced deprecated highlight_style in TUI - Cross-compilation for macOS ARM64 - Idiomat | Medium | 4/1/2026 |
| v5.0.0 | ## What's New in v5.0.0 ContribAI fully rewritten in Rust โ 100% feature parity with Python v4.1.0. ### Performance vs Python v4 | Metric | Python v4 | Rust v5 | |---|---|---| | Startup | ~800ms | ~5ms | | Speed | 1x | ~10โ50x | | Binary | needs runtime | ~4.5 MB | | Memory | ~120 MB | ~8 MB | ### 21 CLI Commands `run` `hunt` `patrol` `target` `analyze` `solve` `stats` `status` `leaderboard` `models` `templates` `profile` `cleanup` `notify-test` `system-status` `web-server` `schedule` `mcp- | Medium | 3/31/2026 |
| v5.1.0 | ## What's New in v5.1.0 ### ๐ฅ๏ธ Interactive TUI (\contribai interactive\) Full terminal UI built with **ratatui + crossterm** โ browse PRs, repos, and run operations without leaving the terminal. | Tab | Description | |---|---| | Dashboard | Stats (total PRs, merge rate, repos analyzed) + recent PRs | | PRs | Full scrollable history with โ
/โ/๐ก status colors | | Repos | Per-repo merge rate display | | Actions | 14 CLI commands with keyboard preview | **Keys:** \Tab\/\1-4\ switch tabs ยท \j\/\k | Medium | 3/31/2026 |
| v4.1.0 | ### Added - **Antigravity MCP Integration**: ContribAI MCP server now works with Antigravity IDE (Google Gemini) in addition to Claude Desktop โ configure via `mcp_config.json` for native tool access to all 14 GitHub operations - Documented MCP setup for both Claude Desktop and Antigravity IDE ### Changed - **PR Title Format**: Removed emoji prefixes from generated PR titles for a cleaner, more professional appearance (`"Quality: fix race condition"` instead of `"โจ Quality: fix race condition" | Medium | 3/28/2026 |
| v4.0.0 | ### Added - **Repo Intelligence Layer** (`contribai/analysis/repo_intel.py`): Profiles target repos before contributing โ analyzes merged PR patterns, identifies high-value issues, tracks review speed, and injects intelligence into LLM prompts for focused contributions - **Smart Dedup (PR History Injection)**: Past PR titles injected directly into analysis prompts with "DO NOT REPEAT" instruction โ prevents rediscovering already-fixed bugs - **Issue-First Hunt Strategy** (`_hunt_issues_globally | Medium | 3/28/2026 |
| v3.0.6 | ## What's New ### ๐๏ธ Auto-close Linked Issues - When a PR is closed (CI failure or maintainer reject), linked issues are now automatically closed - Extracts \Closes/Fixes/Resolves #N\ from PR body and closes each issue - New \close_issue()\ method in GitHubClient with \state_reason: not_planned\ - Works in both pipeline (CI failure) and patrol (maintainer close detection) ### ๐ซ SKIP_DIRECTORIES Filter - Skip findings in low-value directories: \examples\, \example\, \samples\, \demo\, \docs\, | Medium | 3/27/2026 |
| v3.0.5 | ## v3.0.5 โ Critical Security & Reliability Fixes ### Critical Fixes - **Webhook signature bypass**: FastAPI returned HTTP 200 instead of 403 on invalid signatures โ now returns 403 via JSONResponse - **RetryMiddleware re-entry bug**: Shared mutable index caused retries to skip downstream middlewares โ now uses immutable per-call index - **Context compressor wrong kwarg**: `system_prompt=` corrected to `system=` for LLM providers ### High Priority Fixes - **Webhook payload size bypass**: Missi | Medium | 3/27/2026 |
| v3.0.4 | ## v3.0.4 โ Security Hardening, Notification Retry & Docs ### Security - **Timing attack fix**: API key verification now uses `hmac.compare_digest()` for constant-time comparison - **Payload size limit**: Webhook endpoint validates `Content-Length` header (10 MB max) to reject oversized payloads ### Improved - **Notification retry**: Failed sends now retry with exponential backoff (3 attempts: 1s, 2s, 4s) - **MCP client timeout**: `StdioMCPClient` timeout is now configurable via constructor pa | Medium | 3/27/2026 |
| v3.0.3 | ## v3.0.3 - PR Permission Pre-check ### New Feature - **Collaborator-only repo detection**: Automatically detects repos that restrict PRs to collaborators before running expensive analysis/generation pipeline (~30min of LLM calls saved per skipped repo). ### How It Works (3 layers) 1. **Pre-check** (\_check_pr_permissions()\ in pipeline): Calls GitHub permission API + checks \llow_forking\ repo setting BEFORE starting analysis 2. **Both paths covered**: Check inserted in both \_process_repo\ | Medium | 3/27/2026 |
| v3.0.2 | ## v3.0.2 - Critical Generation Fix ### Bug Fix - **[CRITICAL] Fix AttributeError in _validate_changes()**: The validation method was treating FileChange Pydantic objects as raw dicts (using .get()), causing AttributeError: 'FileChange' object has no attribute 'get' on **every successful generation**. This was the root cause of all 'generated: 0' failures across all repositories. ### Improvements - **Tuned self-review prompt**: Changed from strict ('Does it follow good coding practices?') to | Medium | 3/27/2026 |
| v3.0.1 | ## ๐ง v3.0.1 โ Code Generation Quality Improvements ### Bug Fixes - **Robust JSON extraction** โ Handle LLM responses with extra text, markdown fences, or multiple JSON objects (fixes 'Extra data' parse errors) - **String-aware bracket validator** โ Skip brackets inside strings and comments to prevent false validation rejections ### Improvements - **Retry with feedback** โ Automatically retry generation once with error context when parse/validate fails - **Fuzzy matching (difflib)** โ Add Sequ | Medium | 3/27/2026 |
| v3.0.0 | ## ๐ v3.0.0 โ Major Release ### New Features #### F1: EventBus Pipeline Integration - 10 event types emitted throughout the pipeline (PIPELINE_START/COMPLETE, ANALYSIS, GENERATION, PR_CREATED, HUNT_ROUND, MEMORY, ERROR) - Automatic JSONL file logging via `FileEventLogger` - Full observability into pipeline execution #### F2: ContextCompressor in Analyzer - Token-budget compression applied to fetched source files before LLM analysis - Automatic truncation + signature extraction for large file | Medium | 3/27/2026 |
| v2.8.1 | ## What's New ### Human-in-the-Loop Review Gate - New `review_gate.py` โ Rich terminal UI for reviewing contributions before PR creation - Shows finding details, generated code changes, and commit message in formatted panels - User chooses: **[y]es** (create PR), **[n]o** (reject), **[s]kip** - Configurable via `pipeline.human_review: true` in config.yaml - Default: `false` (backward compatible โ preserves autonomous behavior) **Full Changelog**: https://github.com/tang-vu/ContribAI/compare/v2 | Medium | 3/26/2026 |
| v2.8.0 | ## What's New ### AgentScope-Inspired Patterns Adopted best practices from [AgentScope](https://github.com/agentscope-ai/agentscope) (Alibaba DAMO, 8.5k stars). #### LLM-Driven Context Compression - `ContextCompressor.summarize_with_llm()` โ structured summaries via LLM instead of brute-force truncation - Template: Task Overview, Current State, Important Discoveries, Context to Preserve - Graceful fallback to truncation if LLM unavailable #### Auto-Load/Save Working Memory - Pipeline auto-lo | Medium | 3/26/2026 |
| v2.7.0 | ## What's New ### 4 Architecture Patterns (inspired by DeerFlow, SWE-agent, OpenHands, Codex) - **P1: History Compression** โ `ContextCompressor` reduces token usage by truncating, extracting Python signatures, and summarizing findings compactly - **P2: Working vs Archival Memory** โ `working_memory` table with TTL-based hot/cold context separation for smarter cross-repo learning - **P3: Event-Stream Architecture** โ `EventBus` with 15 typed events, async subscribers, and JSONL file logging fo | Medium | 3/26/2026 |
| v2.6.0 | ## What's New in v2.6.0 ### ๐ MCP Server (PR #3 โ @Hoang130203) - **14 MCP tools** exposed via stdio for Claude Desktop integration - GitHub Read: search_repos, get_repo_info, get_file_tree, get_file_content, get_open_issues - GitHub Write: fork_repo, create_branch, push_file_change, create_pr, close_pr - Safety: check_duplicate_pr, check_ai_policy - Maintenance: patrol_prs, cleanup_forks, get_stats - Resource cleanup on shutdown (try/finally) - Fork delete safety guard (verifies `fork==true` | Medium | 3/26/2026 |
| v2.5.0 | ## What's New in v2.5.0 ### Pipeline Wiring (DeerFlow Architecture) - **MiddlewareChain** now fully wired into pipeline (5 middlewares, config-driven) - **AgentRegistry** wired with 5 agents: Analyzer, Generator, Patrol, Compliance, IssueSolver - **ToolRegistry** wired with GitHub + LLM tools - Generator now receives Memory for repo_preferences injection ### New Features - `contribai system-status` command โ memory stats, recent PRs, GitHub rate limit - `IssueSolverAgent` โ 5th agent wrapping | Medium | 3/25/2026 |
| v2.4.1 | ## Bug Fixes - **Fixed** `summarize_findings()` used `Finding.contribution_type` instead of `Finding.type` โ caused `AttributeError` during hunt mode - **Fixed** SECURITY.md referenced non-existent email domain โ now uses GitHub Issues ## Added - 86 new unit tests for v2.4.0 modules (middleware, skills, registry, protocol) โ **333 total tests, 53% coverage** - `docs/ARCHITECTURE.md` โ detailed architecture documentation with diagrams - `AGENTS.md` โ AI-friendly guide for Copilot, Claude, Codera | Medium | 3/25/2026 |
| v2.4.0 | ## What's New Inspired by [ByteDance DeerFlow 2.0](https://github.com/bytedance/deer-flow), this release introduces 6 architectural improvements: ### ๐๏ธ Architecture - **Middleware Chain** โ 5 built-in middlewares (RateLimit, Validation, Retry, DCO, QualityGate) - **Sub-Agent Registry** โ 4 agents (Analyzer, Generator, Patrol, Compliance) with parallel execution - **Tool Protocol** โ MCP-inspired ToolRegistry with GitHubTool and LLMTool wrappers ### ๐ง Intelligence - **Progressive Skill Load | Medium | 3/25/2026 |
